Uncertainty-Aware Pedestrian Attribute Recognition via Evidential Deep Learning
2026-04-29 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed UAPAR, a new system that recognizes pedestrian attributes like clothing or accessories while also knowing when its guesses might be unreliable. Unlike other methods that always give confident answers, UAPAR uses a special approach called Evidential Deep Learning to measure uncertainty, helping it handle tricky or unclear images better. It also includes a way to focus on important image parts and deals with noisy training data through a careful learning process. Tests show UAPAR works well and can tell when its predictions might be wrong or uncertain.
Pedestrian Attribute RecognitionUncertainty EstimationEvidential Deep LearningCLIP ArchitectureCross-AttentionSpatial Prior MasksEpistemic UncertaintyCurriculum LearningLabel NoisePA100K Dataset
Authors
Zhuofan Lou, Shihang Zhang, Fangle Zhu, Shengjie Ye, Pingyu Wang
Abstract
We propose UAPAR, an Uncertainty-Aware Pedestrian Attribute Recognition framework. To the best of our knowledge, this is the first EDL-based uncertainty-aware framework for pedestrian attribute recognition (PAR). Unlike conventional deterministic methods, which fail to assess prediction reliability on low-quality samples, UAPAR effectively identifies unreliable predictions and thus enhances system robustness in complex real-world scenarios. To achieve this, UAPAR incorporates Evidential Deep Learning (EDL) into a CLIP-based architecture. Specifically, a Region-Aware Evidence Reasoning module employs cross-attention and spatial prior masks to capture fine-grained local features, which are further processed by an evidence head to estimate attribute-wise epistemic uncertainty. To further enhance training robustness, we develop an uncertainty-guided dual-stage curriculum learning strategy to alleviate the adverse effects of severe label noise during training. Extensive experiments on the PA100K, PETA, RAPv1, and RAPv2 datasets demonstrate that UAPAR achieves competitive or superior performance. Furthermore, qualitative results confirm that the proposed framework generates uncertainty estimates that are predictive of challenging or erroneous samples.